← Back to Project App Gallery Illustrative dependency-network model

Schedule sensitivity sandbox

Project Risk Gradient

See where one additional day of protective buffer creates the largest marginal reduction in project risk—and whether that protection moves the commitment date.

This is a transparent learning model, not a calibrated risk forecast. It uses a simple CPM network, an exponential risk response, and finite differences to make the “next best buffer day” visible.

Risk response

Project risk and marginal reduction as the selected task receives buffer.

percentage points / day

Risk-reduction ranking

Higher values indicate more project-risk reduction from the next buffer day.

Explanation for project managers

A risk gradient answers a practical question: where would the next day of contingency buy the most reduction in overall exposure?

  • The ranking shows the marginal reduction from adding one day to each task.
  • A “no finish-date impact” badge means the extra day currently fits inside another path’s float.
  • Critical-path buffers usually move the commitment date, so compare risk yield with schedule cost.
  • Use the sliders to test a small allocation, then reset and compare alternatives.
Advanced explanation for risk and scheduling specialists

The model holds baseline CPM weights constant, adds protective buffer to each task’s baseline total float, and approximates a partial derivative with a one-day forward difference.

Task risk = 100 × exp(−(baseline slack + buffer) × coefficient / 2)

Displayed gradient = Project risk(b) − Project risk(b + 1 day)

The mathematical derivative is negative; the interface reports its positive magnitude as risk reduction. Commitment impact is calculated separately by rerunning the dependency network with buffer added to task duration. This keeps the sensitivity curve stable while still exposing critical-path and topology effects.

Adjust protective buffers

Select a task for the chart, then allocate 0–10 days. Dates below are the recalculated commitment schedule.